Statistical and Machine Learning Projects

Implementation of logistic regression with neural network mindset

• build a general learning architecture including parameter initialization, loss and gradient calculation, and optimization from scratch
• packages: Python numpy, scipy, PIL, h5py
• keywords: logistic regression, image classification, computer vision

Implementation of neural network with one hidden layer

• build a shallow neural network architecture
• implement the parameter initialization, forward and backward propagation, model training and prediction, hyperparameter tuning modules
• packages: Python numpy, sklearn
• keywords: neural network, classification

Implementation of deep neural network from scratch

• implement the forward propagation, loss calculation, and backward propagation modules of a fully connected neural network from scratch
• build a multi-layer model and apply to an image classification task
• packages: Python numpy, PIL, scipy, h5py
• keywords: DNN, image classification, computer vision

Neural network modeling for SIGNS dataset using TensorFlow

• build a neural network using Tensorflow
• apply this model to the SIGNS dataset to recognize numbers from hand signs
• packages: Python tensorflow, h5py
• keywords: image classification, computer vision

Implementation of Convolutional Neural Networks (CNN) from scratch

• implement convolutional and pooling layers for forward propagation of Convolutional Neural Networks (CNN) from scratch
• implement a TensorFlow CNN model and apply to the SIGNS dataset to recognize numbers from hand signs
• packages: Python numpy, tensorflow, PIL, h5py
• keywords: CNN, image classification, computer vision

Implementation of face recognition models

• map face images into encodings using a pretained Inception model
• implement the triplet loss function among the positive, negative, and anchor images
• build models for face verification and face recognition
• packages: Python keras, tensorflow
• keywords: face recognition, computer vision

Implementation of residual networks (ResNets) and application on SIGNS dataset

• build a 50-layer convolutional networks using residual networks (ResNets)
• implement the identity block and convolutional block of a ResNet
• apply this ResNet50 model on the SIGNS dataset of hand signs
• packages: Python keras, tensorflow
• keywords: CNN, ResNet, computer vision

Implementation of Art generation by neural style transfer

• implement the neural style transfer algorithm and generate novel artistic images using this algorithm
• neural style transfer merges a "content" image and a "style" image to create a "generated" image
• build the TensorFlow graph based on the pre-trained VGG-19 model
• build content cost function and style cost function and combine them to get the total cost
• packages: Python tensorflow, PIL, scipy
• keywords: generative AI, computer vision

Implementation of YOLO for car detection and autonomous driving

• implement the You Only Look Once (YOLO) algorithm for object detection
• output filtering using non-max suppression (NMS)
• packages: Python numpy, tensorflow, keras, yad2k, PIL, scipy
• keywords: CNN, computer vision, object detection, autonomous driving

Word vector representation using GloVe

• similarity measurement between words based on the pre-trained GloVe vectors
• implement word analogy prediction, debiasing (neutralization and equalization) algorithms
• packages: Python numpy
• keywords: NLP, word embedding

Build an Emojifier using word vector representation and LSTM

• convert words to word vector representations using pre-trained GloVe embeddings
• further feed word embeddings into an LSTM to account for word ordering
• packages: Python emoji, keras
• keywords: NLP, word embedding

Implementation of Recurrent Neural Networks (RNN) from scratch

• implement the forward and backward pass for a basic Recurrent Neural Network (RNN) model from scratch
• implement the forward and backward pass a Long Short-Term Memory (LSTM) model from scratch
• packages: Python numpy
• keywords: NLP, RNN, LSTM

Neural machine translation using LSTM with attention

• build a neural machine translation (NMT) model to translate human-readable dates into machine-readable dates
• implement the attention step, and the pre-attention and post-attention LSTMs
• visualization of the attention weights
• packages: Python tensorflow
• keywords: NLP, LSTM, attention, machine translation

RNN LSTM music generation

• build a model to learn musical patterns and then use this model to generate music similar to the given jazz music
• packages: Python keras
• keywords: RNN, LSTM, music generation, generative AI

RNN LSTM text generation character level language model

• build a character-level text generation RNN model to synthesize new text
• implement the gradient clipping and sampling techniques
• generate dinosaur names according to a training set of actual dinosaur names using a RNN model
• generate Shakespeare poems after training on a collection of Shakespearian poems using a LSTM model
• packages: Python numpy, keras
• keywords: NLP, RNN, LSTM, text generation, generative AI

Trigger word detection for speech recognition

• build a audio spectrograms dataset for speech recognition using data synthesis technique
• build a speech recognition architecture including convolutional, GRU, and dense layers
• packages: Python pydub, keras
• keywords: speech recognition, data synthesis

A simulation study of statistical power in linear regression

• perform a simulation study to demonstrate the factors that affect statistical power in linear regression
• specifically, investigate the effects of sample size, signal strength, and noise level
• visualization and explanation of how power is affected by these factors
• packages: R scales
• keywords: statistical power analysis, linear regression

Linear regression prediction of breast cancer patient overall survival: modeling building and assumption testing

• build a linear regresion model for survival prediction using various model building techniques such as AIC model search, Box-Cox response transformation
• collinearity checking, and diagnostics of model assumptions including residual analysis, Q-Q plot, Breusch-Pagan test, and Shapiro-Wilk normality test
• Kaplan-Meier survival probability and Cox proportional hazards regression
• packages: R tidyverse, survival, lm, lmtest
• keywords: survival analysis, linear regression

Predict the housing prices in Ames: linear regression models and tree-based regression models

• build a linear regression model and a tree-based model for housing price prediction
• compare different regularization methods for the linear regression model: L1-norm penalty (Lasso), L2-norm penalty (Ridge) or a combination of L1-norm and L2-norm penalty (Elastic Net)
• data pre-processing techniques such as one-hot encoding and Winsorization
• packages: R glmnet, xgboost
• keywords: regularization, Elastic Net, Gradient Boosting Machine (GBM)

Walmart Store Sales Forecasting: using SVD in linear regression

• train a linear regression model to predict sales based on historical weekly sales data
• use singular value decomposition (SVD) for feature extraction
• package: R lm
• keywords: sales forecasting, SVD

Movie Review Sentiment Analysis: bag-of-words, n-gram, and logistic regression

• construct the vocabulary based on "bag-of-words" and "n-gram" strategy, build the document-term matrix (DTM), and use a logistic regression model with Lasso regularization to trim the vocabulary
• process the input movie review text to embed the text into matrix
• train a cross-validation logistic regression model with Ridge regularization to predict the sentiment
• package: R text2vec, glmnet, proc
• keywords: NLP, word embedding, sentiment analysis

Clinical risk modeling

• logistic regression based risk modeling for diabetic retinopathy
• decision tree and random forests for risk modeling
• risk model evaluation using C-index
• hyperparameter grid search
• missing data processing: imputation
• model interpertation: SHAP analysis
• packages: Python sklearn, shap
• keywords: healthcare

Clinical survival analysis

• Kaplan-Meier analysis and Log-Rank test
• Cox Proportional Hazards modeling
• Harrell's C-index for Cox model evaluation
• random survival forests
• VIMP score for model interpertation
• packages: Python lifelines, r2py, R randomForestSRC
• keywords: healthcare

Clinical treatment effect analysis

• evaluate the results of randomized control trial (RCT)
• T-learner: learning treatment risk and control risk
• evaluation metric: C-statistic-for-benefit
• packages: Python lifelines
• keywords: healthcare

---

Machine Learning / Deep Learning Projects outside this repo

PyTorch based Image classifier and deployment using AWS Elastic Kubernetes Service (EKS)

MovieLense recommendation system and web application deployment

Deep Learning projects for healthcare applications

see this repo here

---

Note that html files can be viewed by adding http://htmlpreview.github.io/? to the front of the url. For example:

Power analysis simulation:
http://htmlpreview.github.io/?https://github.com/delongmeng/projects/blob/master/power_analysis_simulation.html